What is Hypothesis?
A falsifiable prediction written before an experiment: changing X will move metric Y by a stated direction. Hypotheses force teams to name the mechanism, which makes both wins and flat results interpretable.
Related terms
A/B Testing
Comparing two or more variants on randomized audiences to measure causal impact on a chosen metric. Valid tests need a written hypothesis, sufficient sample, and guardrail metrics so wins do not hide damage elsewhere.
Randomization
Assigning users to test variants by chance so groups differ only randomly. Sound randomization is what lets an A/B result claim causation instead of correlation with pre-existing user differences.
Guardrail Metrics
Secondary metrics watched during experiments to catch harm the primary goal misses: latency, error rate, support contacts, unsubscribe rate. A test that lifts conversion while spiking unsubscribes fails its guardrails.
Autocapture
Automatic collection of clicks, pageviews, and form interactions without per-element instrumentation. Fast to start but noisy: retroactive data comes with vague names and mystery events that explicit tracking would have labeled.
Explicit Events
Hand-defined tracking calls placed at meaningful product moments: signup completed, checkout started, invite sent. Explicit events cost engineering time upfront and pay back in clean names and trustworthy funnels.
Event Taxonomy
The naming contract for analytics: object-action conventions, required properties, and ownership per event. Taxonomies prevent the analytics swamp where three teams log the same signup three different ways.